What is a walled garden in advertising?

What defines one

Not size, and not being proprietary. A walled garden is defined by one party controlling all four of: the inventory, the buying interface, the measurement, and the audience data.

When one company holds all four, an advertiser cannot verify what they received independently, cannot compare it against another platform on equal terms, and cannot take their accumulated audience understanding elsewhere.

What it costs an advertiser

• Self-reported performance. The party being paid produces the numbers — see biased attribution.
• No like-for-like comparison. Metrics are defined differently, so cross-platform evaluation requires assumptions rather than arithmetic.
• No portability. Audiences and learnings do not leave.
• Unilateral terms. Pricing, policy and format change on the platform's schedule.

Why they win anyway

Because they work. Walled gardens have the best data, the most attention and, frequently, genuinely the best performance. Refusing to buy them on principle costs an advertiser more than the transparency is worth.

The workable posture is neither refusal nor dependency. Buy them for the reach; maintain an independent measurement read so the performance claim is not the only evidence; and keep meaningful spend on independent supply so the capability still exists when terms change.

The AI version

AI surfaces have every structural advantage the previous generation had, plus one more: the inventory is generated by their own model, in their own product, and cannot be reached anywhere else. There is no equivalent audience to buy elsewhere and no arbitrage.

That makes the consolidation question sharper than it was for social or search — see AI walled gardens — and it is the reason open protocols such as AdCP are being pushed now rather than after the market settles.